 |
BorlandTalk.com Borland discussion newsgroups
|
| View previous topic :: View next topic |
| Author |
Message |
Michael Fullerton Guest
|
Posted: Mon May 17, 2004 9:13 pm Post subject: Comparing voice files with Delphi? |
|
|
Does anyone have any suggestions on how I could use Delphi to compare
two different files (in my case simple speech files) that would
indicate how similar they are to each other? Are there any Bayesian
filter type libraries available?
Thanks in advance
___
Free Project Time Tracker for Delphi Developers
http://www.cyber-matrix.com/pcdelphi.html
|
|
| Back to top |
|
 |
Boian Mitov Guest
|
Posted: Tue May 18, 2004 2:15 am Post subject: Re: Comparing voice files with Delphi? |
|
|
Well, I don't have a Bayesian filter, but I can offer you filters FFT
etc., as a basis for potential further processing. Please visit
www.mitov.com for more details. You can use the products for free for
non commercial purposes. I hope they will help. If you need the
AudioLab, I can send you a prerelease install.
With best regards,
Boian Mitov
Michael Fullerton wrote:
| Quote: | Does anyone have any suggestions on how I could use Delphi to compare
two different files (in my case simple speech files) that would
indicate how similar they are to each other? Are there any Bayesian
filter type libraries available?
Thanks in advance
___
Free Project Time Tracker for Delphi Developers
http://www.cyber-matrix.com/pcdelphi.html
|
|
|
| Back to top |
|
 |
Rene Tschaggelar Guest
|
Posted: Tue May 18, 2004 9:21 am Post subject: Re: Comparing voice files with Delphi? |
|
|
Michael Fullerton wrote:
| Quote: |
Does anyone have any suggestions on how I could use Delphi to compare
two different files (in my case simple speech files) that would
indicate how similar they are to each other? Are there any Bayesian
filter type libraries available?
|
You're aware what kind of task you're at ?
Usually data is compared for similarity with crosscorrelation.
Speech is a bit more complex in the aspect that it cannot be considered
to be "one piece" but of sequences. Therefore you cannot
fouriertransform a sentence and compare it with another. You rather
have to compare wavelets.
Without prior knowledge, don't expect results within a few months.
It becomes even harder with background noise or when something
should be understood.
Rene
|
|
| Back to top |
|
 |
Michael Fullerton Guest
|
Posted: Tue May 18, 2004 3:13 pm Post subject: Re: Comparing voice files with Delphi? |
|
|
"Rene Tschaggelar" <ee123123wqe12 (AT) sd242323 (DOT) com> wrote:
| Quote: | Michael Fullerton wrote:
Does anyone have any suggestions on how I could use Delphi to compare
two different files (in my case simple speech files) that would
indicate how similar they are to each other? Are there any Bayesian
filter type libraries available?
You're aware what kind of task you're at ?
Usually data is compared for similarity with crosscorrelation.
Speech is a bit more complex in the aspect that it cannot be considered
to be "one piece" but of sequences. Therefore you cannot
fouriertransform a sentence and compare it with another. You rather
have to compare wavelets.
|
At this point what would be compared are simple human speech sounds
(ah, ooh, ee) not even words.
| Quote: | Without prior knowledge, don't expect results within a few months.
It becomes even harder with background noise or when something
should be understood.
|
Background noise is not an issue and I just need a comparison of how
similar the person's sound is with the saved sound.
___
Free Project Time Tracker for Delphi Developers
http://www.cyber-matrix.com/pcdelphi.html
|
|
| Back to top |
|
 |
Boian Mitov Guest
|
Posted: Tue May 18, 2004 4:07 pm Post subject: Re: Comparing voice files with Delphi? |
|
|
The package I have also has a wavelet, but you may as well check on
Torry www.torry.net . There was a set of components ( at least one ),
for neuron networks. It looks like a good candidate. You can record
multiple time a voice from a single person, and then use SignalLab
www.mitov.com to get the FFT over time. Feed the output to a neuron
network, and teach it to recognize the phrases.
With best regards,
Boian Mitov
Michael Fullerton wrote:
| Quote: | "Rene Tschaggelar" <ee123123wqe12 (AT) sd242323 (DOT) com> wrote:
Background noise is not an issue and I just need a comparison of how
similar the person's sound is with the saved sound.
___
Free Project Time Tracker for Delphi Developers
http://www.cyber-matrix.com/pcdelphi.html
|
|
|
| Back to top |
|
 |
Rene Tschaggelar Guest
|
Posted: Tue May 18, 2004 4:35 pm Post subject: Re: Comparing voice files with Delphi? |
|
|
Michael Fullerton wrote:
| Quote: | "Rene Tschaggelar" <ee123123wqe12 (AT) sd242323 (DOT) com> wrote:
Michael Fullerton wrote:
Does anyone have any suggestions on how I could use Delphi to
compare two different files (in my case simple speech files) that
would indicate how similar they are to each other? Are there any
Bayesian filter type libraries available?
You're aware what kind of task you're at ?
Usually data is compared for similarity with crosscorrelation.
Speech is a bit more complex in the aspect that it cannot be
considered to be "one piece" but of sequences. Therefore you cannot
fouriertransform a sentence and compare it with another. You rather
have to compare wavelets.
At this point what would be compared are simple human speech sounds
(ah, ooh, ee) not even words.
Without prior knowledge, don't expect results within a few months.
It becomes even harder with background noise or when something
should be understood.
Background noise is not an issue and I just need a comparison of how
similar the person's sound is with the saved sound.
|
Still...
Try to say a sentence multiple times the same way, perhaps with
a day in between.
What is done is compare a word with a set of recordings of
this word. Then with some expertise you may find the word to be
examined to be within the set.
Rene
|
|
| Back to top |
|
 |
Boian Mitov Guest
|
Posted: Tue May 18, 2004 5:29 pm Post subject: Re: Comparing voice files with Delphi? |
|
|
I think a neuron network can help.
Rene Tschaggelar wrote:
| Quote: | Michael Fullerton wrote:
Still...
Try to say a sentence multiple times the same way, perhaps with
a day in between.
What is done is compare a word with a set of recordings of
this word. Then with some expertise you may find the word to be
examined to be within the set.
Rene
|
|
|
| Back to top |
|
 |
Rene Tschaggelar Guest
|
Posted: Tue May 18, 2004 8:56 pm Post subject: Re: Comparing voice files with Delphi? |
|
|
A neuron network is the last straw when all ideas are exhausted.
By throwing a vast cpu power in a rather uncorrelated way at the
problem you'll be stuck for month trying all sort of configurations
with all sort of parameters while still not learning. Neuron networks
tend to keep their secrets for themselves. So unless you have a job
at a research institute and don't like summer wheather, stay away
from these neuron networks.
Rene
Boian Mitov wrote:
| Quote: | I think a neuron network can help.
Rene Tschaggelar wrote:
Michael Fullerton wrote:
Still...
Try to say a sentence multiple times the same way, perhaps with
a day in between.
What is done is compare a word with a set of recordings of
this word. Then with some expertise you may find the word to be
examined to be within the set.
|
|
|
| Back to top |
|
 |
Jan Mitrovics Guest
|
Posted: Tue May 18, 2004 9:39 pm Post subject: Re: Comparing voice files with Delphi? |
|
|
Boian Mitov wrote:
| Quote: | I think a neuron network can help.
|
Neural networks are a sience of their own. You will need a lot of
experience, before you can make effiecient use of them.
If you want to give it a try, there is a component set of neural
networks available at (Help file in Russian...):
http://www.basegroup.ru/download/neuralbase.en.htm
There are also some shareware libraries available at Torry, etc.
If you want to first get a feeling and some background on neural nets,
this might be interesting for you:
http://www-ra.informatik.uni-tuebingen.de/software/JavaNNS/welcome_e.htm
l
If the interest is on classification and statistics on a broader range,
the SDL component suite for Delphi might be interesting:
http://www.lohninger.com
I would initially not start with a neural network, but with PCA / LDA
projections and a Mahalonobis distance or something similar.
Jan
|
|
| Back to top |
|
 |
Boian Mitov Guest
|
Posted: Tue May 18, 2004 9:57 pm Post subject: Re: Comparing voice files with Delphi? |
|
|
Hi Rene,
It was just an idea . I don't use them, but I think they have a
good potential for the particular task. In any case what he is trying to
achieve is not straight forward, and the neuron networks have their
advantages in this area. Indeed this is what they have been designed
for. One of them on the Torry's page has a OCR demo that resembles some
of the stuff needed in this case. With my signal processing components
and the modified OCR code, he may be able to get what he is trying to
achieve, however it will not be easy.
With best regards,
Boian
Rene Tschaggelar wrote:
| Quote: | A neuron network is the last straw when all ideas are exhausted.
By throwing a vast cpu power in a rather uncorrelated way at the
problem you'll be stuck for month trying all sort of configurations
with all sort of parameters while still not learning. Neuron networks
tend to keep their secrets for themselves. So unless you have a job
at a research institute and don't like summer wheather, stay away
from these neuron networks.
Rene
|
|
|
| Back to top |
|
 |
Jan Mitrovics Guest
|
Posted: Tue May 18, 2004 10:10 pm Post subject: Re: Comparing voice files with Delphi? |
|
|
Rene Tschaggelar wrote:
| Quote: |
A neuron network is the last straw when all ideas are exhausted.
|
Using them this way is asking for trouble.
| Quote: | By throwing a vast cpu power in a rather uncorrelated way at the
problem you'll be stuck for month trying all sort of configurations
with all sort of parameters while still not learning.
|
More problematic in my experience is that they may also fool you by
overfitting if you are not aware of how to use them correctly.
They may be powerfull in the hands of very experienced people, but they
lead the unexperienced directly into chaos.
| Quote: | Neuron networks
tend to keep their secrets for themselves.
|
There are ways to get information from the network itself, but it's not
straight forward and needs careful planning. Another problem is how to
integrate preexisting knowledge into them (unlike Bayesian classifiers).
| Quote: | So unless you have a job
at a research institute and don't like summer wheather, stay away
from these neuron networks.
|
LOL. Yep, that's the way it is!
Jan
|
|
| Back to top |
|
 |
Boian Mitov Guest
|
Posted: Tue May 18, 2004 10:14 pm Post subject: Re: Comparing voice files with Delphi? |
|
|
Well, you have convinced me guys . I will stay away from them.
Well, then again, I like to play games, so I may, just may be will touch
them sometime in the future, but surely not now .
Take care,
Boian
Jan Mitrovics wrote:
| Quote: |
So unless you have a job
at a research institute and don't like summer wheather, stay away
from these neuron networks.
LOL. Yep, that's the way it is!
Jan
|
|
|
| Back to top |
|
 |
Jan Mitrovics Guest
|
Posted: Tue May 18, 2004 10:39 pm Post subject: Re: Comparing voice files with Delphi? |
|
|
Boian Mitov wrote:
| Quote: | Well, you have convinced me guys . I will stay away from them.
Well, then again, I like to play games, so I may, just may be will
touch them sometime in the future, but surely not now . Take
care, Boian
|
Boian,
Your Open Wire stuff is propably quite appropriate for neural networks.
It might actually be fun to play with them (if you have some spare time
and no project schedule that can be screwed up...)
Jan
|
|
| Back to top |
|
 |
Boian Mitov Guest
|
Posted: Tue May 18, 2004 10:39 pm Post subject: Re: Comparing voice files with Delphi? |
|
|
Hi Jan,
Thanks! That's what I had in mind . I am working on 2 projects 1
Delphi and 1 C++ ( A pure C++ OpenWire version for a particular
customer. It is designed to compile on embedded devices as well, and may
be sent to the outer space if the project succeeds ), plus I am
working on the last component of AudioLab prior prerelease version.
There is some VideoLab work under way as well, so the neuron networks
will have to wait a bit . ( Did I mention that the OpenWire editor
work may be resurrected later this year ? )
With best regards,
Boian
Jan Mitrovics wrote:
| Quote: |
Boian,
Your Open Wire stuff is propably quite appropriate for neural networks.
It might actually be fun to play with them (if you have some spare time
and no project schedule that can be screwed up...)
Jan
|
|
|
| Back to top |
|
 |
Boian Mitov Guest
|
Posted: Tue May 18, 2004 10:42 pm Post subject: Re: Comparing voice files with Delphi? |
|
|
BTW: I am doing some research work in the direction of computer vision,
and the neuron networks may come in handy
Boian
Jan Mitrovics wrote:
| Quote: |
Boian,
Your Open Wire stuff is propably quite appropriate for neural networks.
It might actually be fun to play with them (if you have some spare time
and no project schedule that can be screwed up...)
Jan
|
|
|
| Back to top |
|
 |
|
|
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum
|
|